4.5 Discussion
With the free access to Landsat archive data, LULC change dynamics can be
completely explored by extracting LULC information from the temporally dense
and extensive time-serial classification maps. Complexities of LULC change of
urban systems of the Region of Waterloo have been successfully detected. One
major finding has been obtained from the time-serial trajectory analysis and it
shows that LULC change processes of urban area of the Region of Waterloo are
very complex, not simply increasing or decreasing all the way. Taking built-up area
as an example, it experienced dramatic growth over the time period, but the
coverage still had irregular fluctuation up and down during the process. Water
and forest which were not supposed to change too much, also experienced observable fluctuation during this time period.
Apart from the real change, those fluctuations might be resulted from other two
aspects. One is the classification error which cannot be completely eliminated
because of the medium spatial resolution of Landsat data and atmospheric noise.
The other one might be the phenological effects that influence the classification
results. Under this circumstance, the use of long-term dense datasets reveals its
superiority of reducing the impacts caused by those factors. Time-series trajectory
analysis detects the long-term change of complex ecosystems in a macroscopic
view and reducing reliance on one single classification map. For example, it can be
detected that there was an acceleration of growth of urban built-up area of the
Region of Waterloo in 1990s and deceleration in late 2000s. Such valuable information of change complexities are required by environmental researchers and
decision makers. However, it cannot be detected by bi-temporal method or coarsely
multi-temporal method.
4.6 Limitations and Uncertainties
Based on the case study of the Region of Waterloo there are some limitations and
uncertainties of performing change dynamics analysis. From the data perspective,
in order to detect long-term dynamic change of urban area the remote sensing data
are required for sufficiently dense and extensive in time. With long-term record and
free open policy Landsat archive data is the best choice for this study. However,
Landsat data with medium spatial resolution ( 30 m) cannot detect every subtle
object on land surface. Therefore, classification errors should be counted in a study
since it cannot be eliminated. In addition, in this study a classification was
performed on each Landsat image taken from 1984 to 2013 except for 1988,
2004, and 2012. Training samples were selected for each year. In this way, the
quality of classification maps can be guaranteed because the training samples are
sufficient. However, for high dense dataset selecting training samples for each year
was a huge task in this study. The work might become more burdensome when the
82
A. Fu et al.
With the free access to Landsat archive data, LULC change dynamics can be
completely explored by extracting LULC information from the temporally dense
and extensive time-serial classification maps. Complexities of LULC change of
urban systems of the Region of Waterloo have been successfully detected. One
major finding has been obtained from the time-serial trajectory analysis and it
shows that LULC change processes of urban area of the Region of Waterloo are
very complex, not simply increasing or decreasing all the way. Taking built-up area
as an example, it experienced dramatic growth over the time period, but the
coverage still had irregular fluctuation up and down during the process. Water
and forest which were not supposed to change too much, also experienced observable fluctuation during this time period.
Apart from the real change, those fluctuations might be resulted from other two
aspects. One is the classification error which cannot be completely eliminated
because of the medium spatial resolution of Landsat data and atmospheric noise.
The other one might be the phenological effects that influence the classification
results. Under this circumstance, the use of long-term dense datasets reveals its
superiority of reducing the impacts caused by those factors. Time-series trajectory
analysis detects the long-term change of complex ecosystems in a macroscopic
view and reducing reliance on one single classification map. For example, it can be
detected that there was an acceleration of growth of urban built-up area of the
Region of Waterloo in 1990s and deceleration in late 2000s. Such valuable information of change complexities are required by environmental researchers and
decision makers. However, it cannot be detected by bi-temporal method or coarsely
multi-temporal method.
4.6 Limitations and Uncertainties
Based on the case study of the Region of Waterloo there are some limitations and
uncertainties of performing change dynamics analysis. From the data perspective,
in order to detect long-term dynamic change of urban area the remote sensing data
are required for sufficiently dense and extensive in time. With long-term record and
free open policy Landsat archive data is the best choice for this study. However,
Landsat data with medium spatial resolution ( 30 m) cannot detect every subtle
object on land surface. Therefore, classification errors should be counted in a study
since it cannot be eliminated. In addition, in this study a classification was
performed on each Landsat image taken from 1984 to 2013 except for 1988,
2004, and 2012. Training samples were selected for each year. In this way, the
quality of classification maps can be guaranteed because the training samples are
sufficient. However, for high dense dataset selecting training samples for each year
was a huge task in this study. The work might become more burdensome when the
82
A. Fu et al.